WO2022172481A1

WO2022172481A1 - Operation assistance device, operation assistance method, and operation assistance program

Info

Publication number: WO2022172481A1
Application number: PCT/JP2021/026965
Authority: WO
Inventors: 孝朗関合; 鶯仙鄭; 洋人武内
Original assignee: 株式会社日立製作所
Priority date: 2021-02-15
Filing date: 2021-07-19
Publication date: 2022-08-18
Also published as: AU2021427367A9; JP2022124130A; JP7474208B2; AU2021427367A1; AU2021427367B2

Abstract

Provided are an operation assistance device, an operation assistance method, and an operation assistance program, which are capable of providing effective guidance for a control device to operate equipment by using image information received from a plurality of cameras that monitor the state of each part of the equipment. This operation assistance device acquires image data obtained by photographing the equipment by the cameras, and provides guidance on operation of the equipment by an optimal control algorithm using the image data. The operation assistance device is characterized in that the optimal control algorithm quantifies the image data as features and uses the quantified image data, and the image data is obtained by photographing different positions of the equipment at different time points.

Description

Driving support device, driving support method, and driving support program

The present invention relates to a driving support device, a driving support method, and a driving support program that support the operation of various devices, plants, and systems (hereinafter simply referred to as devices).

In recent years, with the technological innovations of ICT (Information and Communication Technology) and IoT (Internet of Things), an environment is being prepared in which high-speed computers, network communications, and large-capacity data storage devices can be used. As the utilization of large amounts of data accumulated in many industrial fields is attracting attention, integration of data collected at local sites such as equipment measurement data, inspection and maintenance data, and systems that manage corporate management and asset information. Therefore, operations to improve key performance indicators are required.

For example, in the field of power generation, there is concern that fluctuations in the amount of power generated due to the increased use of renewable energy such as wind power and solar power will reduce the stability of the power system. sexuality is increasing. In addition, thermal power plants play a role not only as load control but also as a base load power source, and are required to be operated in consideration of operational performance such as efficiency, environmental performance, and operating rate.

In order to improve the operational performance of such thermal power plants, Patent Document 1 discloses a control device that reduces the nitrogen oxide concentration and carbon monoxide concentration, which are environmental performance. According to the technique described in Patent Literature 1, an operation signal is generated by combining a model that simulates the characteristics of a plant and a learning means that learns an optimum operation method for this model. Using this technique, the operating conditions can be moved to optimum values. Here, the operation condition is a value used for generating the operation signal.

JP 2009-128972 A

A thermal power plant measures various state quantities (operating data) at various locations within the thermal power plant, and executes processing using this measurement data. To give a specific example, in a thermal power plant, when ash generated when coal is burned adheres to heat exchangers and furnace walls, the heat transfer characteristics change and the amount of heat absorption decreases. Also, the ash adhering to the heat exchanger is removed by a soot blower (steam injection).

However, it is difficult to grasp the state of ash adhesion only from the various state quantities (operation data) measured at various locations in the thermal power plant. Furthermore, there is a problem that the heat balance changes due to a change in heat transfer characteristics, a decrease in heat absorption, or a temporary decrease in efficiency when steam is used by a soot blower.

On the other hand, cameras have been conventionally used to monitor various devices. For example, if the device is a thermal power plant, the combustion state of the burner is photographed and monitored. However, the conventional camera monitoring monitors the combustion state by paying attention to the time-series changes in the acquired images. In addition, cameras are used to monitor various locations in thermal power plants, but these are only used to monitor the locations where the images were taken, and do not focus on the relationship between images taken at different locations. .

It goes without saying that the technique of applying Patent Document 1 or surveillance cameras can be applied not only to thermal power plants but also to general equipment.

In view of the above, in the present invention, when operating equipment using a control device, there is provided a driving support device that makes it possible to provide effective guidance using image information from a plurality of cameras that monitor the state of each part of the equipment. An object is to provide a support method and a driving support program.

Based on the above, in the present invention, "a driving support device that obtains image data of a device photographed by a camera and provides guidance regarding operation of the device by an optimal control algorithm using the image data, wherein the optimal control algorithm is: A driving support device characterized by using image data as feature quantities in numerical form, and wherein the image data is image data captured by equipment at different locations and at different times.”

In addition, in the present invention, "a driving support method that provides guidance regarding the operation of equipment by means of an optimum control algorithm using image data obtained by photographing the equipment with a camera, wherein the optimum control algorithm digitizes the image data as a feature amount. and the image data is image data photographed at different locations and times by the device.
”.

In addition, in the present invention, there is provided a "driving support program that provides guidance regarding the operation of a device using a plurality of image data obtained by photographing a plurality of locations of the device with a camera. A state recognition program that recognizes the state, a state evaluation program that evaluates the state based on other image data as a numerical feature value and obtains an evaluation value, and a state evaluation program that learns actions to transition to the state that maximizes the evaluation value A driving assistance program characterized by including a learning program."

According to the present invention, effective guidance can be given using image information from a plurality of cameras that monitor the state of each part of the equipment when the equipment is operated by the control device.

For example, by applying the present invention to a thermal power plant, the state of ash adhesion can be grasped based on image data, and the combustion state can be controlled so that the amount of ash adhesion does not increase. That is, as image information from multiple cameras that monitor the state of each part, the image data of the combustion area and the image data of the ash adhesion state are correlated and learned, so that the combustion state will be the desired ash adhesion state. , can control the air flow balance. Also, by locally blowing the soot at the place where the ash adheres, it is possible to eliminate the soot blowing to unnecessary places and reduce the amount of steam consumed. The above can give useful guidance that can improve the efficiency of the boiler plant.

1 is a block diagram illustrating a configuration example of a driving assistance device according to an embodiment of the present invention; FIG. 4 is a flowchart for explaining the operation of the driving assistance device according to the embodiment of the present invention; The figure which shows the example of the driving|running data D11 preserve|saved at driving|running data database DB11. FIG. 4 is a diagram showing an example of image data D12 stored in an image data database DB2; FIG. 4 is a diagram for explaining an example of processing operation of the preprocessing means 40, particularly for image data D12; 4A and 4B are diagrams for explaining the operation of the preprocessing means 40 when the data recording cycles are different; FIG. FIG. 10 is a diagram showing the result of extracting feature amounts from image data D12; 4 is a diagram showing a neural network model as a model included in learning means 70. FIG. The figure which shows the relationship between the input of a neural network model, and an output. FIG. 5 is a diagram showing an example of a result of operating the learning means 70; The figure which shows the structural example at the time of using a coal-fired power plant as the apparatus 19 of FIG. FIG. 3 is a diagram showing an example of image data D12a of the vicinity of the burner 102, which is the combustion area of the boiler 101, photographed by the camera 71a. The figure which shows an example of the image data D12b which image|photographed the vicinity of the heat exchanger 106 of the boiler 101 with the camera 71b. FIG. 5 is a diagram showing an example of feature amounts extracted by processing the image data D12a of the combustion region by the state recognition means 600; FIG. 5 is a diagram showing an example of feature amounts extracted by processing the image data D12b of the heat exchanger 106 by the state evaluation means; 4 is a flowchart for explaining the operation contents of learning means 700 when a coal-fired power plant is used as equipment. The figure explaining the learning result at the time of using a coal-fired power plant as apparatus.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram illustrating a configuration example of a driving support device 20 according to an embodiment of the present invention and related devices. In this embodiment, the driving support device 20 is connected to the control device 18 and the target device 10 including the device 19 to be controlled, and the external device 90 .

The driving assistance device 20 in FIG. 1 is generally configured by a computer device (computer). That is, an arithmetic unit such as a CPU executes various processing functions according to a program for realizing an optimum control algorithm. If the processing functions (optimal control algorithm) in the arithmetic unit are schematically shown, preprocessing means 40, state evaluation means 50, state recognition means 60, learning means 70, action determination means 80, and these processing functions are operated. It can be said that the driving support device operation control means 25 is provided to allow the operation to be performed. Each program described in this embodiment can be distributed to each device via a network, or can be stored in a storage medium and distributed. Details of the processing functions, which are the operations of the respective units in the driving support device 20, will be described with reference to FIG. 2 and subsequent figures. Note that each means described above can also be implemented as hardware. Here, in the present embodiment, each component is expressed as ``means'', but expressions such as ``unit'', ``unit'', etc. are not limited.

The driving support device 20 includes a measurement signal database DB1 and a processing result database DB2 as databases DB. The measurement signal database DB1 includes a driving data database DB11 and an image data database DB12 (DB12a, DB12b, DB12n). Electronic information is stored in each database DB, and the information is stored in a form usually called an electronic file (electronic data). Incidentally, these databases DB may be provided outside the driving assistance device 20 and configured to be connectable via a network.

The driving assistance device 20 also includes an external input interface 21 and an external output interface 22 as interfaces with the outside. The driving operation support device 20 is connected to the target device 10 to which it is applied and the external device 90 via these devices.

The implementation of the driving support device 20 includes each aspect described below. The first is cloudization of the driving support device 20 . This is a configuration in which the driving support device 20 is configured on a public network and can be used by the external device 90 . The second is a mode in which the operating company of the target device 10 operates and manages the driving support device 20 . This is a configuration in which the driving support device 20 is connected to the in-house network of the operating company of the target device 10 and operated and managed by the operating company. The present invention may be according to any of these aspects.

Next, the external device 90 is composed of a computing device (computer). That is, an arithmetic unit such as a CPU executes various processing functions described below according to a program. Also, the external device 90 can be realized as a terminal device, and includes a tablet, a smart phone, a notebook PC, and the like. The external device 90 includes an external input device 91 implemented by a keyboard 92 and a mouse 93 and an image display device 94 . The operator of the target device 10 can operate the external input device 91 based on the information displayed on the image display device 94 to operate the target device 10 .

Also, the target device 10 is composed of a control device 18 and a device 19 . Here, a measurement signal Sg70 is transmitted from the device 19 to the control device 18, and an operation signal Sg80 is transmitted from the control device 18 to the device 19. FIG. The device 19 is provided with a plurality of cameras 71 (71a, 71b, 71n) for capturing images at different locations. The measurement signal Sg70 includes operation data indicating the operation of the device 19 (time-series process value data collected by a sensor provided in the device) and image data captured by a camera. Further, the operation signal Sg80 indicates what kind of signal the control device 18 outputs according to the operation.

The driving support device 20 takes in the external input signal Sg1 and the measurement signal Sg2 via the external input interface 21, and these measurement signals Sg3 are stored in the measurement signal database DB1. The measurement signal Sg3 includes driving data D11 and image data D12, which are stored in the driving data database DB11 and the image data database DB12, respectively. The image data database DB12 manages the image data D12 for each shooting location. In this embodiment, the image data D12a captured by the camera 71a is stored in the a-point image database DB12a, the image data D12b captured by the camera 71b is stored in the b-point image database DB12b, and the image data D12n captured by the camera 71n is stored in the n-point image database DB12n. saved respectively.

The stored image data DB 12 is images captured by fixed-point cameras, drones, underwater drones, and human cameras, and various image data are stored according to the purpose. Moreover, the camera used for photographing may be a high-sensitivity CMOS camera, an infrared camera, a laser camera, or a combination of various cameras.

The preprocessing means 40 acquires the measurement signal Sg4 stored in the measurement signal database DB1, performs data preprocessing as appropriate, and then converts the preprocessed data Sg5 used for state evaluation to the state evaluation means 50. , the preprocessed data Sg6 used for state recognition is transmitted to the state recognition means 60. FIG. The preprocessing means 40 performs correction processing of the measurement signals stored in the measurement signal database, taking into consideration the dead time and delay time in the equipment.

The state evaluation means 50 extracts a feature amount from the preprocessed data Sg5 used for state evaluation, evaluates whether or not this feature amount is a desirable value, and outputs a state evaluation result Sg7. The state evaluation result Sg7 is transmitted to the learning means 70 and the processing result database DB2.

The state recognition means 60 extracts a feature amount from the preprocessed data Sg6 used for state recognition, recognizes the operating state of the target device based on this feature amount, and outputs a state recognition result Sg8. The state recognition result Sg8 is transmitted to the learning means 70, the action determination means 80, and the processing result database DB2.

The feature quantities extracted by the state evaluation means 50 and the state recognition means 60 are the value, size, color, density, temperature, brightness, wavelength, and the value obtained by specifying the object in the image and encoding its contents. rate of change and the like.

The learning means 70 learns an operation method that makes the state evaluation result Sg7 a desired value, and outputs a learning result Sg9. The learning result Sg9 is transmitted to the processing result database DB2. The learning result Sg9 includes information on actions suitable for the current state recognition result. The learning means 70 can be implemented using an optimization algorithm such as reinforcement learning, genetic algorithm, nonlinear programming, etc. However, the present invention does not limit the implementation method of the learning means 70 .

The processing result database DB2 stores the state evaluation result Sg7, the state recognition result Sg8, and the learning result Sg9 obtained as a result of operating the state evaluation means 50, the state recognition means 60, and the learning means 70.

The action determination means 80 refers to the learning result Sg10, determines an action suitable for the current state recognition result Sg8, and outputs an action Sg11. Behavior Sg11 is transmitted to the external output interface 22 .

The external output interface 22 converts the action Sg11 into an operation recommendation signal Sg12 and transmits it to the control device 18 or the image display device 94. As a result, the target device 10 can be directly controlled using the operation recommendation signal Sg12, or the target device 10 can be manually operated with reference to the operation recommendation signal Sg12 displayed on the image display device 94. .

In addition, in the driving assistance device 20 of the present embodiment, an example is shown in which the arithmetic device constituting the computer device and the database DB are provided inside the driving operation assistance device 20 . However, some of these devices may be arranged outside the driving support device 20, and only data may be communicated between the devices.

Also, the database signals stored in each database DB can be displayed on the image display device 94 via the external output interface 22 . Moreover, the values of these signals can be modified by an external input signal Sg1 generated by operating the external input device 91. FIG.

In this embodiment, the external input device 91 is composed of a keyboard 92 and a mouse 93, but any device for inputting data such as a microphone for voice input and a touch panel may be used.

It goes without saying that the present embodiment also includes a method using the driving support device 20. In this embodiment, the guidance target device 10 to which the driving support device 20 is applied is composed of the control device 18 and the equipment 19, but it goes without saying that the device can be implemented as equipment other than this configuration.

FIG. 2 is a flowchart for explaining the operation of the driving support device 20. FIG. This flowchart is realized by the driving support device operation control means 25 operating each arithmetic device. The operation of the arithmetic device can be divided into functions related to learning and functions related to action. The left side of FIG. 2 right is a flow chart for generating an operation recommendation signal for the target device 10 based on the learning result.

First, the operation of the functions related to learning on the left side of Fig. 2 will be explained. In processing step S10, past measurement signals are taken in and stored in the measurement signal database DB1. In processing step S11, the preprocessing means 40 is operated to generate preprocessed data Sg5 used for state evaluation and preprocessed data Sg6 used for state recognition from the measurement signal Sg4. In step Sg12, the state evaluation means 50 and the state recognition means 60 are operated to generate a state evaluation result Sg7 and a state recognition result Sg8. Finally, in processing step S13, the learning means 70 is operated to generate a learning result Sg9. The state evaluation result Sg7, the state recognition result Sg8, and the learning result Sg9 generated in this flowchart are stored in the processing result database DB2.

Next, the operation of the function related to the action on the right side of Fig. 2 will be explained. In processing step S20, the latest measurement signal Sg2 is taken into the measurement signal database DB1 and stored. In processing step S21, the preprocessing means 40 is operated with respect to the latest measurement signal Sg4 to generate post-preprocessing data Sg6 used for state recognition. In processing step S22, the state recognition means 60 is operated to generate a state recognition result Sg8. In processing step S23, action determining means 80 is operated to generate action Sg11. Then, the operation recommendation signal Sg12 is transmitted to the control device 18 or the image display device 94 via the external output interface 22 .

In processing step S24, it is determined whether or not driving assistance should be continued by the driving assistance device 20. If necessary, the process returns to processing step S20. As a method for determining whether or not driving assistance is required to be continued in processing step S24, there is a method in which the operator of the target device 10 inputs information on whether or not driving assistance is required to be continued using the external input device 91, and determination is made according to the content of the information. .

　Figs. 3 and 4 are diagrams for explaining aspects of data stored in the measurement signal database DB1. FIG. 3 shows an example of the operating data D11 stored in the operating data database DB11, and FIG. 4 shows an example of the image data D12 stored in the image data database DB2.

As shown in FIG. 3, the operating data database DB11 stores, for example, time-series data for each data item (item A, item B, item C, . . . ) for each sampling period. Item A is, for example, temperature, item B is flow rate, and item C is pressure. Further, as shown in FIG. 4, the image data database DB2 stores, for example, the temperature distribution measured at a cross section of the device 19 for each sampling period. Operation data and image data of the target device 10 can be displayed on the image display device 94 .

5 and 6 are diagrams for explaining an example of the processing operation of the preprocessing means 40. FIG. First, FIG. 5 particularly relates to image data D12. Here, an outline of a correction method when there are an image shot at point a and an image shot at point b will be described. If the time (dead time and delay time) required for a substance such as fluid to reach point a from point a to point b is Δt12 under operating condition 1, the image captured at point b is corrected forward by Δt12. and then process it. This allows accurate learning of causal relationships. Also, the time required for a substance such as fluid to reach point a from point b (waste time and delay time) is set to Δt34 under operating condition 2, and the delay compensation time is appropriately corrected while judging the operating conditions. It's good.

In the present invention, we are trying to grasp the state of the target device by learning images between at least two points in the upstream and downstream relationship of the movement of fluid etc. in the target device. At this time, instead of comparing images measured at the same time, it is necessary to learn about the image when the fluid was measured on the upstream side and the image when the fluid reached the downstream measurement point. It is preferable to perform delay time compensation in the preprocessing.

FIG. 6 is a diagram for explaining the operation of the preprocessing means 40 when the data recording cycles are different. Here, there are various variations of the data recording cycle, such as once per second, once per day, once per week, and once per inspection cycle (several months or one year). For example, if point a is recorded once per second (real time) and point n is recorded once a week, the average feature value for the period t5-t6 at point a and , with the value of the feature amount at t6 of point n.

Next, the operation of the state evaluation means 50 and the state recognition means 60 will be described. In this process, first, for each of the multiple image data D12 (D12a, D12b, D12n) photographed at multiple locations, the feature amount Ci is extracted from the image data D12, and based on the extraction result, the state evaluation means 50 The state of the imaged device is evaluated, and the state of the imaged device is recognized by the state recognition means 60 . Here, i is a code for identifying the item of the feature amount.
Therefore, the feature amount Ci is time-series information including the acquisition time information of the image data D12, and also the feature amount of the image data D12a, the feature amount of the image data D12b, and the feature amount of the image data D12n. The item of the feature amount is identified by the symbol i.

FIG. 7 shows the result of extracting the feature amount from the image data D12 (D12a, D12b, D12n). As the feature quantity, an object in the image is specified, and its contents are coded values, size, color, density, temperature, brightness, wavelength, and rate of change thereof. Extracted as time-series data. This means that the state of the device indicated by the image data has been re-understood as a numerical feature amount. Note that although the example of FIG. 7 shows an example of object 1, if the multiple image data D12 shot at multiple locations capture multiple objects, the data group of FIG. 7 is generated for each object. ing.

The state evaluation means 50 calculates the evaluation value E using, for example, formula (1).
[Number 1]
E = f(Ci) = ΣWi x Ci (1)
Here, f(Ci) is a function for calculating the evaluation value E. In equation (1), the evaluation value is the sum of products of the weighting parameter Wi and the feature amount Ci. Note that the form of f(Ci) can be arbitrarily set according to the purpose in addition to the formula described above.

Here, both the state evaluation means 50 and the state recognition means 60 extract and use the characteristic amount Ci from the image data D12, but the state of one image (for example, the image data D12a) in the equipment is If there is a relationship that affects the state of the other image (for example, image data D12b), the state recognition means 60 handles the image data D12a on the cause side, and the state evaluation means 50 handles the image data D12b on the result side. should be As a result, the state evaluation means 50 uses the state on the result side as an evaluation value, so that the state on the cause side when optimizing the result can be clearly distinguished and grasped.

Next, the operation of the learning means 70 will be explained. In addition to the driving data D11 accumulated in the driving data database DB11, image data D12 (D12a) obtained at a plurality of locations accumulated in the image data database DB12 (DB12a, DB12b, DB12n) is used for learning by the learning means 70. , D12b, D12n) are used.

In the first embodiment, a case will be described in which the purpose is to learn an action that minimizes the evaluation value E calculated by the state evaluation means 50 . 8(a) and 8(b) are diagrams for explaining the details of the model included in the learning means 70. FIG. The model is constructed by a neural network model as shown in FIG. 8(a), and outputs evaluation values in response to state inputs.

FIG. 8(b) is a diagram showing the relationship between the input and output of the neural network model. According to the neural network model, the input evaluation value can be interpolated to obtain the evaluation value for any state.

FIG. 8(c) is an example showing the result of operating the learning means 70. FIG. In the example of FIG. 8(c), when the current state is in region A, the action is determined to increase the state value, and when the current state is in region B, the action is determined to decrease the state value. . By changing the state in this way, the evaluation value becomes a minimum value, and the evaluation value can be improved.

In the above description, it is needless to say that the model to be installed in the learning means 70 and the learning means 70 can be constructed by a neural network model or other techniques, as shown in FIGS. 8(a), 8(b), and 8(c). stomach.

According to the first embodiment, the information of the camera images taken at different locations of the equipment is digitized and provided for learning. image information can be used to provide effective guidance.

More specifically, according to the present invention, it is possible to obtain from image data information that cannot be obtained from driving data measured by a sensor alone. This is due to the fact that it was used for In particular, by using images taken at a plurality of different locations, there is an effect that it is possible to acquire, for example, the relationship between the upstream and downstream sides of the fluid, or the phenomenon of cause and effect by learning.

In the second embodiment, a case will be described in which the operation support device and the operation support method described in the first embodiment are applied to a coal-fired power plant. FIG. 9 shows a configuration example when a coal-fired power plant is used as the equipment 19 in FIG. First, with reference to FIG. 9, the mechanism of power generation by a coal-fired power plant will be briefly described.

In FIG. 9, on the wall surface of a boiler 101 that constitutes a coal-fired power plant, which is equipment 19, pulverized coal, which is fuel obtained by finely pulverizing coal in a mill 134, primary air for pulverized coal transportation, and 2 air for combustion adjustment. A plurality of burners 102 are provided for supplying secondary air. Then, the pulverized coal supplied through this burner 102 is burned inside the boiler 101 .

As shown in the drawing, the structure of the burners 102 is arranged in a plurality of stages in the vertical direction of the wall surface of the boiler 101, and in each stage a plurality of burners are arranged in a row. With the burner structure and arrangement shown in FIG. 9, inside the boiler 101, the pulverized coal is burned from the front surface (hereinafter referred to as "can front") and the back surface (hereinafter referred to as "can rear") of the boiler wall surface. By improving the burner combustion balance before and after the can, the heat recovery effect of the boiler is improved, and the thermal efficiency of the plant is also improved.

The pulverized coal and the primary air are led to the burner 102 from the pipe 139 and the secondary air from the pipe 141, respectively. The primary air is guided from the fan 120 to the pipe 130, and branches into a pipe 132 passing through the air heater 104 installed on the downstream side of the boiler 101 on the way and a pipe 131 bypassing the air heater 104. do. However, the pipe 133 arranged downstream of the air heater 104 merges again and is led to the mill 134 installed upstream of the burner 102 for producing pulverized coal. The primary air passing through the air heater 104 is heated by exchanging heat with combustion gas flowing down the boiler 101 . Together with this heated primary air, the primary air bypassing the air heater 104 conveys the differentiated coal ground in the mill 134 to the burner 102 .

The mills 134 are arranged so as to correspond to each burner stage (four units in FIG. 9), and supply pulverized coal and primary air to the burners constituting each stage. That is, when the supply of coal is to be reduced, such as when the output of power generation is reduced, the mill can be stopped and the burners can be stopped for each burner stage. In the mill 134, the rotation speed of the mill is adjusted in consideration of the combustibility of the boiler 101 so as to obtain pulverized coal with a desired particle size depending on the properties of the coal to be used. Also, the coal stored in the coal bunker 136 is guided to the coal feeder 135 via the coal conveyor 137, and the coal feeder 135 adjusts the supply amount. It is then fed to mill 134 via coal conveyor 138 .

In addition, the boiler 101 is provided with an after air port 103 that introduces air for two-stage combustion into the boiler 101 . Air for second-stage combustion is led from the pipe 142 to the after air port 103 . In the boiler 101 shown in FIG. 9, the air introduced from the pipe 140 using the fan 121 is similarly heated by the air heater 104 . After that, the secondary air is branched into a pipe 141 for secondary air and a pipe 142 for an after-air port, and led to the burner 102 and the after-air port 103 of the boiler 101, respectively. The flow rate of the air supplied to the burner 102 and the after air port 103 can be adjusted by operating air dampers (not shown) installed in the

pipes

141 and 142, respectively.

High-temperature combustion gas generated by burning pulverized coal inside the boiler 101 flows downstream along a path inside the boiler 101 and is supplied with water by a heat exchanger 106 arranged inside the boiler 101. and heat exchange to generate steam. After that, the exhaust gas flows into the air heater 104 installed on the downstream side of the boiler 101 , heat is exchanged by the air heater 104 , and the temperature of the air supplied to the boiler 101 is raised. As a result, the exhaust gas that has passed through the air heater 104 is subjected to exhaust gas treatment (not shown) and then released into the atmosphere from the chimney.

In addition, the feed water circulating through the heat exchanger 106 of the boiler 101 is supplied to the heat exchanger 106 via the feed water pump 105, is superheated in the heat exchanger 106 by the combustion gas flowing down the boiler 101, and is converted into high-temperature, high-pressure steam. Become. In addition, although the number of heat exchangers is one in this embodiment, a plurality of heat exchangers may be arranged. The high-temperature, high-pressure steam generated by the heat exchanger 106 is guided to the steam turbine 108 via the turbine governor 107 , and the steam turbine 108 is driven by the energy of the steam to generate electricity by the power generator 109 .

Here, in the coal-fired power plant of this embodiment, various measuring instruments are arranged to detect state quantities that indicate the operating state of the plant. Operation data D1, which are measurement signals of the coal-fired power plant acquired from measuring instruments arranged in the coal-fired power plant, are stored in the operation database DB11 in the measurement signal database DB1 shown in FIG. In the case of a coal-fired power plant, as a sensor (measuring instrument) for acquiring the operation data D1, there is, for example, one shown in FIG. That is, a temperature measuring instrument 151 that measures the temperature of the high-temperature, high-pressure steam supplied from the heat exchanger 106 to the steam turbine 108, a pressure measuring instrument 152 that measures the pressure of the steam, and an amount of electric power generated by the generator 109 is measured. There is a power generation output measuring instrument 153 and the like.

Also, feed water generated by cooling steam by a condenser (not shown) of the steam turbine 108 is supplied to the heat exchanger 106 of the boiler 101 by the feed water pump 105. Measured by 150. A measurement signal of a state quantity relating to the concentration of components contained in exhaust gas, which is combustion gas discharged from the boiler 101 , is measured by a concentration measuring device 154 provided downstream of the boiler 101 . Components contained in the exhaust gas include nitrogen oxides (NOx), carbon monoxide (CO), hydrogen sulfide (H ₂ S), and the like.

In addition, there are the following measuring instruments related to the coal supply system. A primary air flow meter 155 for measuring the flow rate of primary air supplied to the mill 134 through the pipe 133, and a coal feeder 135 to measure the amount of coal supplied to the mill 134 through the coal conveyor 138. There is a coal feed meter 156 and a tachometer 157 that measures the number of revolutions of the mill 134 . These are configured to measure the above information for each mill and coal feeder.

That is, in the second embodiment applied to a coal-fired power plant, the following information is used as the operating data D11 accumulated in the operating data database DB11, for example. These are the coal flow rate supplied to the boiler 101, the rotation speed of the mill 134, and the primary and secondary air supplied to the boiler 101, which are the state quantities of the target device 10, which is a coal-fired power plant, measured by the above measuring instruments. The flow rate, the feed water flow rate supplied to the heat exchanger 106 of the boiler 101, the temperature of the steam generated in the heat exchanger 106 of the boiler 101 and supplied to the steam turbine 108, the feed water supplied to the heat exchanger 106 of the boiler 101 These are the feedwater pressure, the gas temperature and gas concentration of the exhaust gas discharged from the boiler 101, the exhaust gas recirculation flow rate for recirculating part of the exhaust gas discharged from the boiler 101 to the boiler 101, and the like.

The above is an example of the operation data D11 stored in the operation database DB11 in the measurement signal database DB1. A plurality of sets of image data D12 are saved.

In the second embodiment applied to a coal-fired power plant, the image data D12 stored in the image data database DB12 includes, for example, image data D12a captured by the camera 71a that captures the combustion area on the wall surface of the boiler 101, and the heat exchanger 106 of the boiler 101. There are image data D12b captured by the camera 71a, image data D12c captured by the piping of the boiler 101, and the like. In general, a large number of measuring instruments or cameras other than those shown in FIG. 9 are arranged in the coal-fired power plant, but the illustration is omitted here.

Here, I would like to clarify the problems with coal-fired power plants. In a coal-fired power plant, when ash generated when coal is burned adheres to heat exchangers and furnace walls, the heat transfer characteristics change and the amount of heat absorption decreases. Also, the ash adhering to the heat exchanger is removed by a soot blower (steam injection). A soot blower is designed to inject steam into each heat exchanger.

Regarding this problem, analysis has conventionally been performed using the operating data D11 detected by sensors at various locations in the thermal power plant, but it is difficult to grasp the state of ash adhesion from the operating data D11 alone. In addition, heat balance changes due to changes in heat transfer characteristics and a decrease in heat absorption. The use of steam by the sootblower is temporarily less efficient.

From this, when applying the operation support device 20 of the present invention to a coal-fired power plant, when processing the image data D12b of the ash adherence of the heat exchanger 106, it is taken when the heat exchanger 106 is not adhered with ash. It is preferable to evaluate the amount of ash adhered on the basis of the printed image.

Also, it is possible to improve the accuracy by matching the information grasped from the image data D12 and the information grasped from the process data D11. That is, by correcting the degree of contamination estimated from the process data D11 based on the state of adhesion of ash ascertained from the image data D12b, the accuracy of estimating the degree of contamination can be improved.

The driving support device 20 of the present invention grasps the state of ash adhesion based on the image data D12b. Then, the combustion state can be controlled so that the ash adhesion amount does not increase. That is, the image data D12a of the combustion area and the image data D12b of the ash adhesion state are learned in association with each other, and the balance of the air flow rate can be controlled so as to achieve the desired combustion state of the ash adhesion state. In addition, since the image data D12b can grasp the position where the ash adheres, by blowing the soot locally, it is possible to eliminate the soot blower to unnecessary places and reduce the amount of steam consumed. The above effect can improve the efficiency of the boiler plant.

Next, an example of image data D12 when a coal-fired power plant is used as equipment will be described. FIG. 10a is an example of image data D12a of the vicinity of the burner 102, which is the combustion area of the boiler 101, photographed by the camera 71a. From this image, the vicinity of the burners 102 arranged vertically and horizontally on the wall surface of the boiler 101 has the highest temperature of 1050 degrees, and the outer edge indicates the areas of 1000 degrees and 950 degrees as appropriate. The temperature distribution in the combustion region, such as direction, can be grasped. Although the temperature distribution is described in this embodiment, information such as color, density, brightness, and wavelength may be included in the image data.

Also, FIG. 10b is an example of image data D12b of the vicinity of the heat exchanger 106 of the boiler 101 photographed by the camera 71b. From this image, the amount of ash adhering to the heat exchanger 106 can be grasped.

This example focuses on the upstream combustion region and the downstream heat exchange region in the thermal fluid, and has a causal relationship between the cause and effect of the upstream state affecting the downstream state. . By arranging the cameras at these positions, it is possible to photograph the combustion state and heat exchange state. In addition, it is possible to grasp the above causal relationship.

Next, an example of the feature amount extracted from the image data D12 when using a coal-fired power plant as the equipment will be described. FIG. 11a is an example of the feature amount extracted by processing the image data D12a of the combustion area by the state recognition means 60. FIG. The average value of the temperatures on the left side and the right side of the can is extracted as a feature quantity and saved as time-series data.

FIG. 11b is an example of feature amounts extracted by processing the image data D12b of the heat exchanger 106 by the state evaluation means 50. FIG. The amount of ash adhered in the heat exchangers 106 such as the primary superheater (1SH) and the secondary superheater (2SH) is extracted as feature quantities and stored as time-series data.

FIG. 12 is a flow chart explaining the operation contents of the learning means 70 when a coal-fired power plant is used as equipment. First, in processing step S30, the load plan is captured. Here, the load plan is a plan for the power output (load) of the thermal power plant, and is determined so as to satisfy the power demand.

Next, in processing step S31, an operation plan change proposal is created. Here, the operation plan includes the timing of injecting the soot blower, the set value of the air amount, the type of coal (coal type), and the like.

In processing step S32, the amount of ash adhered when the operation plan change proposal is implemented is estimated. In this processing step S32, the state corresponding to the operation plan (the state grasped by the state recognition means 60 when processing the past image data) and the ash adhesion amount are associated and learned, and based on the result, the ash adhesion amount is determined. presume.

In processing step S33, efficiency is estimated based on the amount of ash adhered, and an evaluation value is calculated. Here, the efficiency is calculated according to the amount of ash adhered. Also, the evaluation value is calculated by a function with efficiency as an input, and the higher the efficiency, the higher the evaluation value.

In processing step S34, the pros and cons of the operation plan change proposal created in processing step S31 are learned. In other words, it learns that an operation plan change proposal with a high evaluation value is a good change plan, and an operation plan change proposal with a low evaluation value is a bad change plan. Create a proposal that will be expensive.

In processing step S35, learning end determination is performed. If YES, the learning is terminated, and if NO, the process returns to step S31. For example, when processing steps S30 to S35 are repeated a predetermined number of times, the learning ends.

FIG. 13 is a diagram for explaining learning results when a coal-fired power plant is used as equipment.
In FIG. 13, the horizontal axis indicates time, and the vertical axis indicates, from top to bottom, the load plan for a certain month, the timing of soot blower injection, the two-stage combustion ratio, which is an example of the air amount set value, and the coal type.

The soot blower injection in FIG. 13 shows an example in which the injection area is divided into 4 areas of the can right and can left of the heat exchanger for the primary heat exchanger 1SH and the secondary heat exchanger 2SH. By injecting steam into the heat exchanger at an appropriate timing for each region and removing the ash adhering to the heat exchanger, the heat absorption amount of the heat exchanger can be improved, and the efficiency can be improved. On the other hand, since the soot blower injection uses high-temperature steam in the plant, it is necessary to replenish make-up water (pure water, etc.) after that, and to make high-temperature steam. , temporarily reduces efficiency.

Therefore, it is desirable to divide the soot blower into small areas and perform each area at the appropriate timing, instead of performing the soot blower injection on the entire heat exchanger at once. Also, it is desirable to spray the soot blower to the position of the heat exchanger where the ash adheres. In the driving assistance device 20 of the present invention, it is possible to grasp the position where the ash adheres from the image data, and to blow the soot to that position, which contributes to the efficiency improvement.

In addition, air operation makes it possible to maintain combustion conditions that reduce the amount of ash adhesion and to select coal types that take ash adhesion into consideration, which also contributes to efficiency improvement.

As described above, by using the operation support device 20 of the present invention in a coal-fired power plant, it is possible to improve the efficiency of the plant.

In the third embodiment, a program that a driving support device configured using a computer should have in the ROM will be described.

Programs to be provided in the ROM may correspond to each processing function shown in FIG. It has a state evaluation program that processes image data of the state of the downstream side of the fluid and obtains an evaluation value, and a learning program that learns using the image processing results and obtains the operation method of the equipment as an action.

The driving support program, which is composed of these individual programs, is used as appropriately specialized by the target device.

Sg1: external input signal, Sg2: measurement signal, Sg3: measurement signal, Sg4: measurement signal, Sg5: preprocessed data used for state evaluation, Sg6: preprocessed data used for state recognition, Sg7: state evaluation result , Sg8: state recognition result, Sg9: learning result, Sg10: learning result, Sg11: action, Sg12: operation recommendation signal, Sg70: measurement signal, 71: camera, Sg80: operation signal, 10: target device, 18: control device , 19: equipment, 20: driving support device, 21: external input interface, 22: external output interface, DB1: measurement signal database, DB11: driving data database, DB12: image data database, DB2: processing result database, 40: front Processing means 50: State evaluation means 60: State recognition means 70: Learning means 80: Action determination means 90: External device 91: External input device 92: Keyboard 93: Mouse 94: Image display device

Claims

A driving support device that obtains image data of a device captured by a camera and provides guidance on driving the device by an optimum control algorithm using the image data,
The driving support system, wherein the optimum control algorithm uses the image data as a feature quantity after being digitized, and the image data is image data captured by the device at different locations and times.
The driving support device according to claim 1,
The optimal control algorithm is an algorithm that recognizes a state based on the image data, calculates an evaluation value based on the image data, and learns behavior for transitioning to a state with the maximum evaluation value,
The driving assistance device, wherein the image data used for recognizing the state and the image data used for calculating the evaluation value are image data photographed at different positions of the device.
The driving support device according to claim 2,
The optimal control algorithm learns using data obtained by preprocessing the image data,
The driving assistance device, wherein the preprocessing corrects the image data in consideration of dead time and delay time in the device.
The operation support device according to claim 2 or 3, wherein the equipment is a boiler plant,
Acquiring the image data of the combustion section on the upstream side of the thermal fluid and the downstream side of the thermal fluid of the boiler plant, and providing guidance on the operation of the boiler plant so that the conditions on the downstream side of the thermal fluid have desired characteristics. A driving support device characterized by:
The driving support device according to claim 4,
The driving assistance device, wherein the guidance is boiler parameters or a sootblower operation method.
The driving support device according to claim 4,
The state is recognized using image data of the combustion section on the upstream side of the thermal fluid of the boiler plant, the evaluation value is calculated using the image data on the downstream side of the thermal fluid, and as an action in reinforcement learning A driving support device characterized in that the determined guidance is parameters of a boiler or an operation method of a sootblower.
The driving support device according to claim 6,
When the image data on the downstream side of the thermal fluid is image data of ash adhesion on the heat exchanger, the amount of ash adhesion is evaluated based on an image taken when ash is not adhered to the heat exchanger. driving support device.
A driving assistance method that provides guidance regarding the operation of the equipment by an optimum control algorithm using image data of the equipment captured by a camera,
The driving support method, wherein the optimum control algorithm uses the image data as a feature quantity after being digitized, and the image data is image data captured by the equipment at different locations and times.
The driving support method according to claim 8,
The optimal control algorithm is an algorithm that recognizes a state based on the image data, calculates an evaluation value based on the image data, and learns behavior for transitioning to a state with the maximum evaluation value,
A driving support method, wherein the image data used for recognizing the state and the image data used for calculating the evaluation value are image data photographed at different positions of the device.
The driving support method according to claim 9,
The optimal control algorithm learns using data obtained by preprocessing the image data,
The driving support method, wherein in the preprocessing, the image data is corrected in consideration of dead time and delay time in the device.
The operation support method according to claim 9 or 10, wherein the equipment is a boiler plant,
Acquiring the image data of the combustion section on the upstream side of the thermal fluid and the downstream side of the thermal fluid of the boiler plant, and providing guidance on the operation of the boiler plant so that the conditions on the downstream side of the thermal fluid have desired characteristics. A driving assistance method characterized by:
The driving support method according to claim 11,
The driving support method, wherein the guidance is boiler parameters or a sootblower operation method.
The driving support method according to claim 11,
The state is recognized using image data of the combustion section on the upstream side of the thermal fluid of the boiler plant, the evaluation value is calculated using the image data on the downstream side of the thermal fluid, and as an action in reinforcement learning A driving support method, wherein the determined guidance is a boiler parameter or a soot blower operation method.
The driving assistance method according to claim 13,
When the image data on the downstream side of the thermal fluid is image data of ash adhesion on the heat exchanger, the amount of ash adhesion is evaluated based on an image taken when ash is not adhered to the heat exchanger. driving support method.
A driving assistance program that provides guidance on driving the equipment using a plurality of image data obtained by photographing a plurality of locations of the equipment with a camera,
A state recognition program for recognizing a state based on the image data as a numerical feature amount, a state evaluation program for evaluating a state based on the other image data as a numerical feature amount and obtaining an evaluation value, A driving assistance program comprising a learning program for learning actions for transitioning to a state in which the evaluation value is maximized.